A few days ago we've post a question about "servlet engine hung arround 200
users".
We have investigate. In fact the servlet engine (in a separate VM from the
ejb server, see config below) is waiting about many simultaneous
multi-threaded call to InitialContext.lookup(on session bean) when more than
200 simultaneous users are connected !
I precise that each threads manipulate a different InitialContext instance
(so don't need synchronize...). And jprobe threadanalyser doesn't detect any
deadlock on the webServer nor on the ejbServer...
The servlet engine is hung in 60 to 150 simultaneous lookup() that never
return.
The problem appear when the web server do remote rmi call for all ejb (see
config below).
==> Is anyone have already seen this kind of problem in lookup ? What's
wrong on the rmi port or rmi registry ??
thank's for any help,
Stan.
PS:
1. We stress our appli with WebLoad (throw http request from 4
different client host), on NT with following config:
- WebServer: orion1.3.8+jdk1.3 on a PIII 700MHz - 512 Mo RAM
and all ejb call are remote call to following ejb server.
with orion-application.xml:
<orion-application >
<ejb-module remote="true" .... /> // for all ejb-module
........
</orion-application>
and rmi.xml:
<rmi-server>
<server host="EjbServer" port="23791" .../>
........
</rmi-server>
- EjbServer: orion1.3.8+jdk1.3 on a PIII 500MHz - 256 Mo RAM
with rmi.xml:
<rmi-server port="23791">
........
</rmi-server>
- DataBase: oracle8i ...
2. if we stress the application without remote call for the ejb's
everything is OK for more than 500 simultaneous users !!
- Web+EjbServer: orion1.3.8+jdk1.3 on a PIII 700MHz - 512 Mo RAM
with orion-application.xml:
<orion-application >
<ejb-module remote="false" .... /> // for all ejb-module
........
</orion-application>
and rmi.xml:
<rmi-server>
........
</rmi-server>
- DataBase: oracle8i ...
